<i>Background/Aims:</i> Self-reported history of stroke has been questioned in the elderly due to the high prevalence of cognitive impairment. We tested the validity of response to a stroke questionnaire versus clinical diagnosis of stroke among elderly people with and without cognitive impairment. <i>Methods:</i> Community-dwelling participants to the phase 1 Canadian Study of Health and Aging were screened for self-reported stroke. Physician-diagnosed stroke was set as the gold standard. The positive predictive value (PPV), sensitivity and specificity were determined. <i>Results:</i> 1,536/ 1,659 (93%) participants aged 65 years and over had stroke information from both sources. Among stroke positive responders, the PPV was 81% overall: 76% for cognitively normal, 84% for cognitively impairment with no dementia (CIND), and 82% for demented. Among stroke diagnosed by physicians, history of stroke was reported by 38% cognitively normal, 54% CIND, and 55% demented. The specificity was over 97% in all cognitive categories. <i>Conclusion:</i> Among community-dwelling elderly people, any cognitive impairment did not imply inaccurate self-reported history of stroke. High prevalence of stroke and frequent contacts with health services among cognitively-impaired elderly may increase the awareness of stroke symptoms and signs. Stroke increases the risk of developing dementia in both cognitively normal and CIND, and efforts to accomplish stroke prevention are justified, especially in these categories.